| Large-scale volumetric fracturing is one of the key technologies for the development of tight oil and gas and shale oil and gas.The evaluation of post-fracturing involves the intersection of mechanics,geology,mathematics and computer science.In this dissertation,by studying the bottom hole pressure of tight reservoirs and the multi-stage fracturing formation pressure distribution of horizontal wells,a parallel computational algorithm of formation pressure distribution based on GPU and a new method based on machine learning technology to identify multi-stage fracturing horizontal well parameters are proposed.Exploring the application of big data has important guiding significance for the development and improvement of big data in the field of oil and gas reservoirs.At the same time,considering the actual tight oil and gas development,the longer the horizontal well,the larger the fracturing scale and the higher the output,but the development cost and water consumption also increase sharply.Through the post-fracturing effect evaluation and capacity prediction,the optimization study of multi-stage fracturing of horizontal wells is carried out,which has important application value for the efficient economic development of tight oil and gas and shale oil and gas.The main research innovations are as follows:1.An unstable seepage model for multi-stage fracturing horizontal wells is established.Through the study of the state equation of tight oil and gas,the continuity equation and the constitutive relation of rock,the seepage equation of multi-stage fractures in horizontal wells is established.The deterministic solution of the seepage model is obtained by the mathematical function analysis methods such as source function theory,three-dimensional eigenvalue method and orthogonal transformation.Finally,through the Laplace transform and the numerical inversion of Stehfest,the bottom hole pressure solution of the multi-stage fracturing in the horizontal well considering the wellbore storage effect and the skin factor is obtained.2.A parallel computing method for multi-stage fracturing of horizontal wells based on GPU is proposed.In this dissertation,a mathematical model for multistage hydraulically fractured horizontal wells(MFHWs)in tight oil and gas reservoirs is derived by considering the variations in the permeability and porosity of tight oil and gas reservoirs that depend on formation pressure and mixed fluid properties,and introducing the pseudo-pressure;analytical solutions were presented using the Newman superposition principle.The CPU-GPU asynchronous computing model was designed based on the CUDA platform,and the analytic solution was decomposed into infinite summation and integral forms for parallel computation.Implementation of this algorithm demonstrates that computation speed increases by almost 80 times,which meets the requirement for real-time calculation of the formation pressure of MFHWs.3.Two new methods of hybrid machine learning are proposed to identify the parameters of MFHW.First,the GPU parallel program is used to calculate bottom hole pressure for multi-stage fractured horizontal wells to obtain bottom hole pressure data.Then,import the data obtained by GPU parallel calculation into the neural network model for training.In the training process,the RBF neural network parameters and SVM parameters are optimized by using the optimization ability of the global optimal solution of the PSO algorithm,and the required calculation model is obtained.Finally,use the remaining data to test the resulting model.The test results show that the performance of the proposed model is better than other models,with the highest correlation coefficient,the lowest mean and absolute error,which proves that the hybrid machine learning model can be effectively applied to horizontal well parameter identification.The work in this dissertation shows that the method of multi-stage fractured horizontal well parameter identification based on a hybrid machine learning model has great potential in petroleum reservoir description,can improve the accuracy of reservoir physical parameter prediction,and is important for petroleum exploration and production.4.A sampling method based on Latin hypercube is proposed to generate high quality model training samples.Deep machine learning technology has a large demand for the amount of data in training samples.The data samples currently available in oil fields are generally not satisfactory.Combining the field measured data and the interpretation results,the Latin hypercube sampling is used to generate the samples needed for the training of the neural network model.It not only makes up for the shortage of the measured sample size,but also ensures the spatial uniformity of the parameter cluster and greatly improves the prediction accuracy of the PSO-RBF neural network model.5.The analysis software developed based on the seepage theory is used for data interpretation and analysis of multi-stage fractured horizontal wells in tight oil fields,and the calculation results of the software are cross-compared with the prediction results obtained by the maching learning model.The results show that both the analysis software and the maching learning model proposed in this dissertation can explain the formation parameters quickly and accurately,and give the capacity prediction. |
PDF Full Text Request